May 16 – 21 , 2021, Dagstuhl Seminar 21201

Serverless Computing


Cristina Abad (ESPOL – Guayaquil, EC)
Ian T. Foster (Argonne National Laboratory – Lemont, US)
Nikolas Herbst (Universität Würzburg, DE)
Alexandru Iosup (VU University Amsterdam, NL)

For support, please contact

Dagstuhl Service Team


Dagstuhl Report, Volume 11, Issue 4 Dagstuhl Report
Aims & Scope
List of Participants
Dagstuhl's Impact: Documents available


Serverless computing holds a significant promise for the modern, digital society. For the past seven decades, our society has increasingly required ever-cheaper, ever-more convenient, and ever-faster computing technology. In the late-1950s, leasing time on an IBM 701 cost $15,000 per month ($135,000 in 2020 dollars). Today, we can lease many times this computing power for mere pennies but need to be careful about the actual cost of doing so. Cloud computing, that is, the utility providing IT as a service, on-demand and pay-per-use, is a widely used computing paradigm that offers large economies of scale and promises extreme environmental efficiency. Born from a need to make cloud computing services more accessible, fine-grained, and affordable, serverless computing has garnered interest from both industry and academia. In our vision, serverless computing can meet this need, but to do this it will have to overcome its current status of emergent technology or risk its demise.

Cloud computing is already an established technology. Today, more than three-quarters of the US and European companies, and many private individuals, use cloud computing services. The serverless market is blooming and has already exceeded $200 billion in 20203. The cost of one hour on a cloud computer leased on-demand can be lower than a cent and all the major cloud providers offer inexpensive access to diverse and state-of-the-art hardware. However cheap, cloud computing still poses daunting operational challenges to software professionals, in particular, how to manage the selection, operation, and other aspects of using cloud infrastructure (in short, servers). Correspondingly, it poses significant challenges to systems designers and administrators, related to keeping the cloud infrastructure efficient and sustainable.

An emerging class of cloud-based software architectures, serverless computing, focuses on providing software professionals the ability to execute arbitrary functions with low or even no overhead in server management. Serverless computing leverages recent developments in the miniaturization of software parts through microservice-based architectures, in the operationalization of small self-contained execution units through containers, and in their integration in service models such as Function-as-a-Service (FaaS). Truly, serverless is more [12]. Early research successes [6, 15, 17, 18, 22] complement numerous industrial applications [9], from business-critical to scientific computing, from DevOps to side-tasks. Already, IT spending on serverless computing should exceed $8 billion per year, by 2021.

However promising, serverless computing has yet to mature and presents many hard, open challenges. There are numerous signs and reports [11, 14] that serverless computing poses critical challenges in software engineering, parallel and distributed systems operation, and performance engineering [10]. For example, software engineering could help overcome challenges in the developer experience [23], including testing, tooling, functionality, and training and education. The systems side requires, among others, new approaches for deployment, monitoring, and general operation, and also specific advances in security, cost predictability, and life-cycle management for cloud functions. Performance engineering raises many hard aspects, such as performance optimization, engineering for cost-efficiency, and various forms of fast online scheduling. These combined challenges are distinctive from the general challenges of cloud computing, for example, because the fine-grained, often event-driven nature of serverless computing typically requires approaches that are lightweight and able to respond without delay.

The goal of the seminar is to combine the views of a diverse and high-quality group of researchers spanning three disciplines: software engineering, parallel and distributed systems, and performance engineering. The Dagstuhl Seminar will be a catalyst. Attendees discussed the open challenges and opportunities of serverless computing for the next decade, with a focus on at least the following crucial aspects and questions:

  • Envision serverless systems and applications in the next decade. How to leverage the freedom from operational concerns? How to overcome the challenge and enjoy the benefits of fine granularity?
  • How to properly engineer serverless software and systems? What are the emerging architectural patterns for serverless systems and applications? How to test and debug serverless systems and applications?
  • How to characterize, model, and analyze serverless systems and applications? How to understand the diverse serverless workloads?
  • How to manage the resources used in serverless operations? How to schedule and orchestrate in this environment? How to manage specific application classes, such as computer vision, enterprise workflows, HPC, DevOps?
  • How to deploy and manage the full lifecycle of serverless applications? How to add ML-capabilities to feedback loops? How to break through the operational silos?
  • How to support privacy, security, dependability, and other desirable operational properties for serverless applications and systems?
  • Beyond computer systems, how to consider serverless systems and applications from a holistic, cyberphysical perspective?

Core Topics

The seminar focussed on the following key topics related to serverless computing:

Topic 1. Design decisions for serverless systems, platforms, and ecosystems. As the serverless model is increasingly being adopted in industry [9], the challenges of properly designing these systems and the platforms on which they run are becoming more apparent. These challenges include important problems [10], such as: how to reduce the serverless overhead added by the platform to the (commonly lightweight) functions representing the business logic of the application (e.g., see [20]), how to ensure proper performance isolation while making efficient use of the shared infrastructure (e.g., see [1]), how to partition the functions [5, 6], and how to properly schedule functions and route requests to these functions (e.g., see [2]), in such a way that the service level objectives (SLO’s) are adequately met, among other important challenges. There is also the question of running serverless workloads alongside conventional applications, e.g., HPC, big data, machine learning. The experiences of the attendees to the seminar, some of which have already started working in these domain and others with established experience in prior technologies from which we may learn and transfer knowledge (e.g., grid computing), will enable us to focus on determining which of these decisions the community should be focusing on, and how to establish adequately prioritized research agendas.

Topic 2. Software engineering of serverless applications, but also systems, platforms, and ecosystems. To increase the domain of application for serverless computing, the functionality it can express needs to become increasingly more complex, which contrasts with the perceived simplicity of the model [23]. What is the trade-off between simplicity and expressiveness? Which composition models can ensure that serverless workflows can be maintained and developed (and updated) long term? Serverless functions should become increasingly interoperable, and applications should become able to leverage the services of any serverless platform [6]. How to make serverless functions vendor-agnostic and how to run serverless applications across cloud federations? Which architectural patterns are useful for serverless applications? How to consider and support the legacy part of serverless applications? The development processes, from the macro-view of how teams coordinate to deliver applications that could operate in an external ecosystem, to the micro-view of how to develop and test a serverless function, will have to consider the new aspects raised by serverless computing. What are effective development processes? What tools and IDE features are needed? What versioning and testing, and what CI/CD protocols should be used? How to evolve legacy software toward serverless-native applications? How to ensure open-source software becomes FAIR software [13]?

Topic 3. Applications and domain requirements for serverless computing. Preliminary studies of serverless applications at large [9] have shown that there is a wide variety of scenarios for which industry and academia are adopting serverless approaches. From business-critical workloads, to automating DevOps, scientific computing, and beyond, the diversity of the applications and domains for which serverless is being applied poses significant challenges when attempting to optimally manage the resources and infrastructure on which these applications depend. It is important to properly understand the variety of these applications and domain requirements, engaging both academia and industry in the discussion. These requirements should relate to various aspects in software engineering, parallel and distributed systems, and performance engineering. For example, a domain-based approach could help increase scalability [3]; considering the structure of packages in composing a deployable serverless application could improve scheduling performance [2]; and serverless functions and architectures should be considered during performance tests [8, 28].

Topic 4. Evaluation of serverless computing systems, platforms, and ecosystems. The performance trade-offs of serverless systems are not yet well understood [28], thus highlighting the importance of proper evaluation and benchmarking of these systems. However, the high level of abstraction and the opaqueness of the operational-side make evaluating these platforms particularly challenging. As recent efforts are starting to focus on this topic [24, 28], it is important to engage the community on an early discussion on the best approaches to tackle this problem. How to understand and engineer the performance of serverless systems? How to translate the findings, when serverless systems are opened to external developers (as platforms) or take part in much larger systems of systems (and even ecosystems)? How to account for parts of the ecosystem being closed-source and even acting as black-boxes? How to identify and even explain the performance bottlenecks such systems, platforms, and ecosystems experience? How to use evaluation results with other performance engineering techniques to control and improve the performance of serverless systems, platforms, and ecosystems?

An important focus of inquiry has recently become prominent in computer systems research: the reproducibility of evaluation results and of experiments in general [19, 21]. Not doing so can result in misleading results [26], and in results that cannot be obtained again [25] sometimes even under identical circumstances and by their original authors [7]. This leads to a possible loss of faith in the entire field [4, 27]. "How to benchmark serverless solutions reproducibly?" is an important question to address with diverse expertise and fresh ideas.

Synopsis and Planned Actions

We would like to thank the Dagstuhl staff and sponsors for this unique seminar opportunity even under the constraints of the pandemic. During the seminar, we had almost 45h of online meetings (not counting sub-meetings): some 9-10h of online meetings each seminar day. Three 3h sessions per day were spread around the clock to allow participation from various timezones. Even under these constraints, we experienced enormous participation and active discussion involvement. In brief, the seminar week was structured as follows:

After each participant presented her/himself to the plenary, we formed four working groups according to the topics above. The discussions were kick-started by four distinguished keynotes, in plenary, with the respective talk abstracts included in this report:

  • "Serverless Predictions: 2021-2030" given jointly by Pedro García López (Universitat Rovira i Virgili – Tarragona, ES) and Bernard Metzler (IBM Research-Zurich, CH)
  • "Developer Experience for Serverless: Challenges and Opportunities" given by Robert Chatley (Imperial College London, GB)
  • "Federated Function as a Service" given jointly by Kyle Chard (University of Chicago, US) and Ian T. Foster (Argonne National Laboratory – Lemont, US)
  • "Characterizing Serverless Systems" given by Mohammad Shahrad (University of British Columbia – Vancouver, CA)

Each of the four working groups held five 3h sessions with their teams, including three 1h one-on-one meetings with the other groups. The four working groups report individually on their outcomes and list identified research challenges. In a consolidation phase, we identified and planned nine focused topics for future joint research among the participants.

Complemented by a Slack workspace for the seminar participants, a focused continuation of discussions beyond the seminar week was enabled: Among others, a discussion initiated and led by Samuel Kounev on the notion of serverless computing, started during the seminar, continued well beyond. We include the outcome of this "panel discussion" in Section 5.1 of this report.

The organizers and participants decided to jointly work toward at least one high-profile magazine article reporting on the seminar outcome and research agenda.

Furthermore, during the seminar the motion was raised to establish a conference series on serverless computing. We see good potential for a new conference on "Serverless Software and Systems" as a cross-community event embracing, at least, the disciplines of software engineering, system engineering, and performance engineering. Working potentially in concert with an existing workshop series in the field, we plan to initiate this step in the coming months. We hope that one day in the future, we can proudly look back and say that this Dagstuhl seminar 21201 was an important trigger event.


  1. Z. Al-Ali, S. Goodarzy, E. Hunter, S. Ha, R. Han, E. Keller, and E. Rozner. Making serverless computing more serverless. In 2018 IEEE 11th International Conference on Cloud Computing (CLOUD), 2018.
  2. G. Aumala, E. Boza, L. Ortiz-Avilés, G. Totoy, and C. Abad. Beyond load balancing: Package-aware scheduling for serverless platforms. In 2019 19th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGRID), 2019.
  3. Alberto Avritzer, Vincenzo Ferme, Andrea Janes, Barbara Russo, André van Hoorn, Henning Schulz, Daniel S. Menasché, and Vilc Queupe Rufino. Scalability assessment of microservice architecture deployment configurations: A domain-based approach leveraging operational profiles and load tests. J. Syst. Softw., 165:110564, 2020.
  4. Monya Baker. Is there a reproducibility crisis? Nature, 533(7604):452-454, 2016.
  5. Edwin F. Boza, Xavier Andrade, Jorge Cedeno, Jorge R. Murillo, Harold Aragon, Cristina L. Abad, and Andres G. Abad. On implementing autonomic systems with a serverless computing approach: The case of self-partitioning cloud caches. Comput., 9(1):14, 2020.
  6. Ryan Chard, Yadu N. Babuji, Zhuozhao Li, Tyler J. Skluzacek, Anna Woodard, Ben Blaiszik, Ian T. Foster, and Kyle Chard. funcx: A federated function serving fabric for science. In HPDC '20: The 29th International Symposium on High-Performance Parallel and Distributed Computing, pages 65-76. ACM, 2020.
  7. Christian S. Collberg and Todd A. Proebsting. Repeatability in computer systems research. Commun. ACM, 59(3):62-69, 2016.
  8. Simon Eismann, Cor-Paul Bezemer, Weiyi Shang, Dusan Okanovic, and André van Hoorn. Microservices: A performance tester's dream or nightmare? In ICPE '20: ACM/SPEC International Conference on Performance Engineering, 2020, pages 138-149. ACM, 2020.
  9. Simon Eismann, Joel Scheuner, Erwin van Eyk, Maximilian Schwinger, Johannes Grohmann, Nikolas Herbst, Cristina L. Abad, and Alexandru Iosup. A review of serverless use cases and their characteristics. Technical Report SPEC-RG-2020-5, SPEC RG Cloud Working Group, May 2020.
  10. Erwin Van Eyk, Alexandru Iosup, Cristina L. Abad, Johannes Grohmann, and Simon Eismann. A SPEC RG cloud group's vision on the performance challenges of FaaS cloud architectures. In Companion of the 2018 ACM/SPEC International Conference on Performance Engineering, ICPE 2018, pages 21-24, 2018.
  11. Erwin Van Eyk, Alexandru Iosup, Simon Seif, and Markus Thömmes. The SPEC cloud group's research vision on FaaS and serverless architectures. In Proceedings of the 2nd International Workshop on Serverless Computing, WOSC@Middleware 2017, Las Vegas, NV, USA, December 12, 2017, pages 1-4, 2017.
  12. Erwin Van Eyk, Lucian Toader, Sacheendra Talluri, Laurens Versluis, Alexandru Uta, and Alexandru Iosup. Serverless is more: From PaaS to present cloud computing. IEEE Internet Comput., 22(5):8-17, 2018.
  13. Wilhelm Hasselbring, Leslie Carr, Simon Hettrick, Heather S. Packer, and Thanassis Tiropanis. From FAIR research data toward FAIR and open research software. it Inf. Technol., 62(1):39-47, 2020.
  14. Joseph M. Hellerstein, Jose M. Faleiro, Joseph Gonzalez, Johann Schleier-Smith, Vikram Sreekanti, Alexey Tumanov, and Chenggang Wu. Serverless computing: One step forward, two steps back. In CIDR 2019, 9th Biennial Conference on Innovative Data Systems Research, Asilomar, CA, USA, January 13-16, 2019, Online Proceedings., 2019.
  15. Scott Hendrickson, Stephen Sturdevant, Tyler Harter, Venkateshwaran Venkataramani, Andrea C. Arpaci-Dusseau, and Remzi H. Arpaci-Dusseau. Serverless computation with openlambda. In 8th USENIX Workshop on Hot Topics in Cloud Computing, HotCloud 2016, Denver, CO, USA, June 20-21, 2016., 2016.
  16. Alexandru Iosup, Catia Trubiani, Anne Koziolek, José Nelson Amaral, Andre B. Bondi, and Andreas Brunnert. Flexibility is key in organizing a global professional conference online: The ICPE 2020 experience in the COVID-19 era. CoRR, abs/2005.09085, 2020.
  17. Eric Jonas, Johann Schleier-Smith, Vikram Sreekanti, Chia-che Tsai, Anurag Khandelwal, Qifan Pu, Vaishaal Shankar, Joao Carreira, Karl Krauth, Neeraja Jayant Yadwadkar, Joseph E. Gonzalez, Raluca Ada Popa, Ion Stoica, and David A. Patterson. Cloud programming simplified: A berkeley view on serverless computing. CoRR, abs/1902.03383, 2019.
  18. Ana Klimovic, YawenWang, Patrick Stuedi, Animesh Trivedi, Jonas Pfefferle, and Christos Kozyrakis. Pocket: Elastic ephemeral storage for serverless analytics. In 13th USENIX Symposium on Operating Systems Design and Implementation, OSDI 2018, Carlsbad, CA, USA, October 8-10, 2018, pages 427--444. USENIX Association, 2018.
  19. Ravi Madduri, Kyle Chard, Mike D'Arcy, Segun C. Jung, Alexis Rodriguez, Dinanath Sulakhe, Eric Deutsch, Cory Funk, Ben Heavner, Matthew Richards, Paul Shannon, Gustavo Glusman, Nathan Price, Carl Kesselman, and Ian Foster. Reproducible big data science: A case study in continuous fairness. PLOS ONE, 14(4):1-22, 04 2019.
  20. Edward Oakes, Leon Yang, Dennis Zhou, Kevin Houck, Tyler Harter, Andrea Arpaci-Dusseau, and Remzi Arpaci-Dusseau. SOCK: Rapid task provisioning with serverlessoptimized containers. In USENIX Annual Technical Conference (USENIX ATC 18), July 2018.
  21. A. V. Papadopoulos, L. Versluis, A. Bauer, N. Herbst, J. Von Kistowski, A. Ali-eldin, C. Abad, J. N. Amaral, P. Tuma, and A. Iosup. Methodological principles for reproducible performance evaluation in cloud computing. IEEE Trans. Software Eng., pages 1-1, 2019.
  22. Qifan Pu, Shivaram Venkataraman, and Ion Stoica. Shuffling, fast and slow: Scalable analytics on serverless infrastructure. In Jay R. Lorch and Minlan Yu, editors, 16th USENIX Symposium on Networked Systems Design and Implementation, NSDI 2019, Boston, MA, February 26-28, 2019, pages 193-206. USENIX Association, 2019.
  23. Mike Roberts. Serverless architectures. serverless.html, 2016. Continuous development of the material.
  24. Joel Scheuner and Philipp Leitner. Function-as-a-service performance evaluation: A multivocal literature review. Journal of Systems and Software, 2020.
  25. Dag I. K. Sjøberg, Jo Erskine Hannay, Ove Hansen, Vigdis By Kampenes, Amela Karahasanovic, Nils-Kristian Liborg, and Anette C. Rekdal. A survey of controlled experiments in software engineering. IEEE Trans. Software Eng., 31(9):733--753, 2005.
  26. Alexandru Uta, Alexandru Custura, Dmitry Duplyakin, Ivo Jimenez, Jan S. Rellermeyer, Carlos Maltzahn, Robert Ricci, and Alexandru Iosup. Is big data performance reproducible in modern cloud networks? In 17th USENIX Symposium on Networked Systems Design and Implementation, NSDI 2020, Santa Clara, CA, USA, February 25-27, 2020, pages 513-527, 2020.
  27. Erik van der Kouwe, Gernot Heiser, Dennis Andriesse, Herbert Bos, and Cristiano Giuffrida. Benchmarking flaws undermine security research. IEEE Secur. Priv., 18(3):48-57, 2020.
  28. Erwin van Eyk, Joel Scheuner, Simon Eismann, Cristina L. Abad, and Alexandru Iosup. Beyond microbenchmarks: The SPEC-RG vision for a comprehensive serverless benchmark. In Companion of the ACM/SPEC International Conference on Performance Engineering (ICPE), page 26-31, 2020.
Summary text license
  Creative Commons BY 4.0
  Cristina Abad, Ian T. Foster, Nikolas Herbst, and Alexandru Iosup


  • Distributed / Parallel / And Cluster Computing
  • Emerging Technologies
  • Software Engineering


  • Serverless computing
  • Cloud continuum
  • Software architecture
  • Systems management
  • Vision


In the series Dagstuhl Reports each Dagstuhl Seminar and Dagstuhl Perspectives Workshop is documented. The seminar organizers, in cooperation with the collector, prepare a report that includes contributions from the participants' talks together with a summary of the seminar.


Download overview leaflet (PDF).

Dagstuhl's Impact

Please inform us when a publication was published as a result from your seminar. These publications are listed in the category Dagstuhl's Impact and are presented on a special shelf on the ground floor of the library.


Furthermore, a comprehensive peer-reviewed collection of research papers can be published in the series Dagstuhl Follow-Ups.